8 Core Nehalem

[JackPack]

Intel is building four 45nm fabs, each of which have 176,000 to 200,000+ sq. feet of clean room space.

AMD's Fab 36 has less than 1/5 this amount of clean room space.

Even if you account for Larrabee and Silverthorne, it doesn't take a whiz to figure out which process node Intel plans to introduce 8-core processors.

 

[JumpingJack]

Originally posted by: Nathelion
Personally, I doubt anything Nehalem will actually be on the market until late 2009 or even 2010. Looking at the feature list, it is such a huge step forward from Core 2 that I can't help but think that it'll take a long time to get the kinks out of it. Intel's got to pull off:
- new memory interface
- monolithic quad core (maybe even octo core?). We've seen AMD have some problems with this one.
- integrated graphics
- hyperthreading on a non-P4 arch (this one might be easy though, i don't know).
- absolutely gargantuan die size with all that cache, 45 nm or not.

That's a pretty tall order for anyone.

Nehalem is currently scheduled for release in Q4 2008, considering that Intel demonstrated a functioning chip, booted and running software Sept 2007, I think the likelihood that they will make this timeframe is quite high.

http://www.xbitlabs.com/news/c...ay/20071015155739.html


Here is the demonstration announcement....
http://www.physorg.com/news109344893.html

Looking to 2008, Otellini made the first public demonstration of Intel's Nehalem processor and said the company is on track to deliver the new processor design in the second half of the year.

http://www.anandtech.com/cpuch...el/showdoc.aspx?i=3102
Here you can see Nehalem in one picture a 2P quad nehalem running, with 16 threads loaded -- part of the SMT.

 

[Dadofamunky]

8 cores. Sheesh. At this rate, the software jocks will be so far behind they'll never catch up.

Now, I could see some awesome developments using this chip for deep packet inspection and threat detection. I work in that area and multi-core CPUs are the only way to go. But we use Cavium chips that run 16 MIPS cores at 600-800 MHz. A new Cavium release will do (I believe) 1 GHz per core.

 

[GFORCE100]

Originally posted by: Nemesis 1

Originally posted by: GFORCE100

Originally posted by: xxceler8
Can anyone confirm details of intel's next chip. I have read so many conflicting articles.

Some say 09 release, some say mid 08 release. If this is indeed a Native octi-core, with IMC, I might scale down my near future purchase in hopes of pulling the trigger on this 8 core.

Nahelm won't be 8 cores until 32nm. Until then Intel will re-enable HT in the Extreme Edition Nahelms.

Unless AMD gets the B3 stepping out, and can scale beyond 3GHz in quantities it can sell them to the retail and OEM channels then Nahelm will be delayed until Q1 2009.

As AMD is really in ill health, expect Intel to slow down with new releases and slip by at least one quarter (i.e. 3 months or more).

Could you show me a link that says 8cores nehalem @32nm only? Please! Thank you!
and a link that says intel will enable H/T in Extreme Edition Nehalems only. AS per your bolded and underlineded statement above qouted.
This Link here says 8 cores @45 no mention of 32nm .

http://www.intel.com/technolog...arch+body_45nm_nehalem

Intel doesn't comment on yet unreleased products.

As far as HT goes, Intel confirms it in the above link you provided in the form of SMT which is a technical term for what HT is.

8 cores at 45nm is possible but not as a mainstream product with the same amount of L2 cache. Also, if you reduce the L2 cache too much, and have too many cores fighting for it, then performance will be degraded significantly. True 8 cores will be at 32nm.

 

[Nemesis 1]

Originally posted by: GFORCE100

Originally posted by: Nemesis 1

Originally posted by: GFORCE100

Originally posted by: xxceler8
Can anyone confirm details of intel's next chip. I have read so many conflicting articles.

Some say 09 release, some say mid 08 release. If this is indeed a Native octi-core, with IMC, I might scale down my near future purchase in hopes of pulling the trigger on this 8 core.

Nahelm won't be 8 cores until 32nm. Until then Intel will re-enable HT in the Extreme Edition Nahelms.

Unless AMD gets the B3 stepping out, and can scale beyond 3GHz in quantities it can sell them to the retail and OEM channels then Nahelm will be delayed until Q1 2009.

As AMD is really in ill health, expect Intel to slow down with new releases and slip by at least one quarter (i.e. 3 months or more).

Could you show me a link that says 8cores nehalem @32nm only? Please! Thank you!
and a link that says intel will enable H/T in Extreme Edition Nehalems only. AS per your bolded and underlineded statement above qouted.
This Link here says 8 cores @45 no mention of 32nm .

http://www.intel.com/technolog...arch+body_45nm_nehalem

Intel doesn't comment on yet unreleased products.

As far as HT goes, Intel confirms it in the above link you provided in the form of SMT which is a technical term for what HT is.

8 cores at 45nm is possible but not as a mainstream product with the same amount of L2 cache. Also, if you reduce the L2 cache too much, and have too many cores fighting for it, then performance will be degraded significantly. True 8 cores will be at 32nm.

1). You said : Intel doesn't comment on yet unreleased products.

My link is an intel statement . Or didn't you read it.

2). You said: Nahelm won't be 8 cores until 32nm. Then Intel will re-enable HT in the Extreme Edition Nahelms.
Again you make it sound like H/T would be re-enabled on Extreme Nehalem only. SMT is for all Nehalem processors. Read your statement.
2a). You do understand How nehalem is being built correct . Its Monolithic:

2. consisting of one piece; solid or unbroken.
3. characterized by massiveness, total uniformity, rigidity, invulnerability, etc.: a monolithic society.
So true 8 cores on Nehalem 45nm would be = to 32nm cores. Minus any small improvements to logic and or the 32nm process. Intel has stated time and time again that Nehalem will be available in many differant configurations. 3 cpu' cores and 1 gpu core along with many other configurations.

You see were it says monolithic society? One society but made up of large numbers of individuals. Monolithic many individual cores linked by quick path to function as a single piece of silicon. Of course the multi leveled shared cache is what makes brings it all together. But it would seem Intel design is a lot differant than AMD K10. AT least I pray it is . Nehalem is really interesting . Nehalem 2 core will hurt AMD really really bad.

2 cores 4 threads . The question all will be asking soon is this.
In a 4 thread apps. Will Nehalem out perform K10. I don't know the ans. I can say this tho.
1). If Nehalem 2core outperforms K10 4 cores . AMD is screwed.
2). AMDs 3 core idea . turns to mush vs. Nehalem 2core version. AMD is going to have to price these babies really low =,< Nehalem 2 cores.

Nehalem is looking to be Intels Perfect Storm!

 

[Phynaz]

Originally posted by: GFORCE100

Originally posted by: Nemesis 1

Originally posted by: GFORCE100

Originally posted by: xxceler8
Can anyone confirm details of intel's next chip. I have read so many conflicting articles.

Some say 09 release, some say mid 08 release. If this is indeed a Native octi-core, with IMC, I might scale down my near future purchase in hopes of pulling the trigger on this 8 core.

Nahelm won't be 8 cores until 32nm. Until then Intel will re-enable HT in the Extreme Edition Nahelms.

Unless AMD gets the B3 stepping out, and can scale beyond 3GHz in quantities it can sell them to the retail and OEM channels then Nahelm will be delayed until Q1 2009.

As AMD is really in ill health, expect Intel to slow down with new releases and slip by at least one quarter (i.e. 3 months or more).

Could you show me a link that says 8cores nehalem @32nm only? Please! Thank you!
and a link that says intel will enable H/T in Extreme Edition Nehalems only. AS per your bolded and underlineded statement above qouted.
This Link here says 8 cores @45 no mention of 32nm .

http://www.intel.com/technolog...arch+body_45nm_nehalem

Intel doesn't comment on yet unreleased products.

As far as HT goes, Intel confirms it in the above link you provided in the form of SMT which is a technical term for what HT is.

8 cores at 45nm is possible but not as a mainstream product with the same amount of L2 cache. Also, if you reduce the L2 cache too much, and have too many cores fighting for it, then performance will be degraded significantly. True 8 cores will be at 32nm.

So at 45nm we'll have faux 8 cores? Something like the fake fur you can buy on QVC?

 

[GFORCE100]

Originally posted by: Nemesis 1

Originally posted by: GFORCE100

Originally posted by: Nemesis 1

Originally posted by: GFORCE100

Originally posted by: xxceler8
Can anyone confirm details of intel's next chip. I have read so many conflicting articles.

Some say 09 release, some say mid 08 release. If this is indeed a Native octi-core, with IMC, I might scale down my near future purchase in hopes of pulling the trigger on this 8 core.

Nahelm won't be 8 cores until 32nm. Until then Intel will re-enable HT in the Extreme Edition Nahelms.

Unless AMD gets the B3 stepping out, and can scale beyond 3GHz in quantities it can sell them to the retail and OEM channels then Nahelm will be delayed until Q1 2009.

As AMD is really in ill health, expect Intel to slow down with new releases and slip by at least one quarter (i.e. 3 months or more).

Could you show me a link that says 8cores nehalem @32nm only? Please! Thank you!
and a link that says intel will enable H/T in Extreme Edition Nehalems only. AS per your bolded and underlineded statement above qouted.
This Link here says 8 cores @45 no mention of 32nm .

http://www.intel.com/technolog...arch+body_45nm_nehalem

Intel doesn't comment on yet unreleased products.

As far as HT goes, Intel confirms it in the above link you provided in the form of SMT which is a technical term for what HT is.

8 cores at 45nm is possible but not as a mainstream product with the same amount of L2 cache. Also, if you reduce the L2 cache too much, and have too many cores fighting for it, then performance will be degraded significantly. True 8 cores will be at 32nm.

1). You said : Intel doesn't comment on yet unreleased products.

My link is an intel statement . Or didn't you read it.

2). You said: Nahelm won't be 8 cores until 32nm. Then Intel will re-enable HT in the Extreme Edition Nahelms.
Again you make it sound like H/T would be re-enabled on Extreme Nehalem only. SMT is for all Nehalem processors. Read your statement.
2a). You do understand How nehalem is being built correct . Its Monolithic:

2. consisting of one piece; solid or unbroken.
3. characterized by massiveness, total uniformity, rigidity, invulnerability, etc.: a monolithic society.
So true 8 cores on Nehalem 45nm would be = to 32nm cores. Minus any small improvements to logic and or the 32nm process. Intel has stated time and time again that Nehalem will be available in many differant configurations. 3 cpu' cores and 1 gpu core along with many other configurations.

You see were it says monolithic society? One society but made up of large numbers of individuals. Monolithic many individual cores linked by quick path to function as a single piece of silicon. Of course the multi leveled shared cache is what makes brings it all together. But it would seem Intel design is a lot differant than AMD K10. AT least I pray it is . Nehalem is really interesting . Nehalem 2 core will hurt AMD really really bad.

2 cores 4 threads . The question all will be asking soon is this.
In a 4 thread apps. Will Nehalem out perform K10. I don't know the ans. I can say this tho.
1). If Nehalem 2core outperforms K10 4 cores . AMD is screwed.
2). AMDs 3 core idea . turns to mush vs. Nehalem 2core version. AMD is going to have to price these babies really low =,< Nehalem 2 cores.

Nehalem is looking to be Intels Perfect Storm!

Yes, and the Intel link is just general information, Intel has an internal policy not to comment on yet unreleased products so no link, human info, or word document stored on an Intel office PC is allowed to be made public before product launch.

Indeed, Nahelm is a monolithic design with the memory controller onboard, as in the AMD K8/K10's.

SMT is not fake, it's all about using free resources on each die to execute an aditional thread in parallel, just like in the Pemtium 4. It's never going to give the performance boost of a real additional physical core albeit will however, offer substancial benefits, especially if the software is coded for it. It's better to have HT than not, even if it's about up to 20% more performance.

AMD is screwed now as it is, no need for Nahelm. If AMD can't get K10 ramped up and at 3GHz or more quick then this = lame profits if at all for Q1 and Q2 and quite frankly AMD is bleeding money now for a good year and a half. I've said time and time again in my other posts over the past year, AMD made a mistake going monolithic with the resources they have and the performance is nothing to write home about. They really got fooled and over-confident with Intel selling Netburst when it's natural fact that Intel has architecture lifespans which means a new architecture every 5 years or so. Given Netburst was released in 2000, it's logical and NYSE sound that Conroe saw the light of day in 2006.

It boils down to people, someone at AMD wasn't seeing the writing on the wall, which first came apparant in early 2004 when it was obvious Intel had to look beyond Netburst if to remain competetive because the heat issues it was having with Prescott. That there should have been a flashing beacon to the minds at AMD, but no...........their ego just grew and instead they focussed their efforts on asking Intel to a duel between their Athlon 64 x2 and Intel's Pentium D.

Business has no sweethearts, you don't catch the rabbit before your competitor, you loose buddy, as easy at that, and as harsh too, for AMD. Their BOD is ill formed if stupid decisions are forced through with the CEO and CTO not having the power to stop these before they cost the company dearly. These decisions are made years ahead, what AMD is witnessing now is down to wrong decisions back in 2003/2004.

You don't get your chief marketing guru jump ship who knows the company inside out if there are no problems or personal tension inside AMD. AMD is most surely in panic mode right now as the sucess of K10 is really hanging on one fine hair now. It's either B3 stepping and 3GHz+ or I dare say shareholders will be getting pretty angry. I would hate to be Hector Ruiz who is obliged to answer their complaints in a public seminar. You don't get paid $1 million for nothing in this world, it's not all nice and rosy as rarely does a company have a 100% success execution rate, especially in firms where the hierarchy is so diverse. Of course if K10 bacame the wonder it was supposed to be then being CEO of AMD would be a delightful job with nothing but applause and some nice caviar at a few dinners. Sadly that's just a dream and since Hector is the one who ultimately signs on the dotted line, if anyone, it will be him and the CTO who get the blame for why K10 is in such a mess. AMD has 65nm problems with it's X2's at above 2.6GHz and magically they thought a bigger die at a massive 285mm2 will be easier to make at 3GHz+. There's some real foolish thinking going on in AMD and I'm somehow confident it's not all down to Hector.

 

[Aberforth]

i doubt they can squeeze 8 cores with 45nm, you need 22nm for that. And what about power? It's just unthinkable. That's why you need two physical cores to run a 8 core box, ,but then even if they make a 8 core cpu i doubt there would be any applications to take real advantage of that.

What we need is a perfect framework for scheduling tasks and new alorithms to manage virtual threads especially in Windows environment. Right now there aren't any applications that take 100% advantage of multi-core processors. The *real* problem is in the software. Software is just pathetically inefficient.

 

[Nemesis 1]

Well lets just take the info intel has given us and assume that Intel is telling the same BS stories AMD put out about K10.

Intel moving to Nehalem is part of intels roadmaps . I hardly think Intel will stop innovating because AMD cann't compete. Nehalem will arrive when Intel said it will be .

Don't look for intel to demo Nehalem @ Spring IDF. Thats going to be great! Amd fanbois will be running around screaming . Nehalem must be in bad shape because Intel didn't show benchies like they did C2D.
Its differant this time . Intel has no competion ! Intel would be silly to demo a Nehalem at spring IDF . Larrobee maybe. Why would intel show a cpu thats going to be anywere from 40%- 80% faster C2D . Let me explain.

C2D runs bench of multi-threaded app. 2 cores 2 threads time to finish .

Nehalem runs multi thread bench 2cores 4 thread Nehalem finishes in 1/2 the time required by C2D.

Thats probably a low performance estimate for multi threaded performance.

As far as single threaded performance thats the one we all would like to know.

 

[BitByBit]

Originally posted by: Nemesis 1
C2D runs bench of multi-threaded app. 2 cores 2 threads time to finish .

Nehalem runs multi thread bench 2cores 4 thread Nehalem finishes in 1/2 the time required by C2D.

Thats probably a low performance estimate for multi threaded performance.

As far as single threaded performance thats the one we all would like to know.

The performance benefit from SMT on Nehalem will only become apparent once the number of threads of a particular program exceed the number of cores. Nehalem will be released in quad core form initially, so for the vast majority of applications, Nehalem will be faster than its predecessor due to superior IPC. Mitosis may change that however.

 

[Extelleron]

Originally posted by: Nemesis 1
Well lets just take the info intel has given us and assume that Intel is telling the same BS stories AMD put out about K10.

Intel moving to Nehalem is part of intels roadmaps . I hardly think Intel will stop innovating because AMD cann't compete. Nehalem will arrive when Intel said it will be .

Don't look for intel to demo Nehalem @ Spring IDF. Thats going to be great! Amd fanbois will be running around screaming . Nehalem must be in bad shape because Intel didn't show benchies like they did C2D.
Its differant this time . Intel has no competion ! Intel would be silly to demo a Nehalem at spring IDF . Larrobee maybe. Why would intel show a cpu thats going to be anywere from 40%- 80% faster C2D . Let me explain.

C2D runs bench of multi-threaded app. 2 cores 2 threads time to finish .

Nehalem runs multi thread bench 2cores 4 thread Nehalem finishes in 1/2 the time required by C2D.

Thats probably a low performance estimate for multi threaded performance.

As far as single threaded performance thats the one we all would like to know.

That's not the way SMT works. Intel's HyperThreading on Netburst CPUs improved performance by around 10-20% by using idle parts of the processor to execute another thread, so 2 instead of 1 or 4 instead of 2. HyperThreading was great, because it gave a noticeable increase in performance with a negligible increase in transistors, meanwhile making general IPC improvements or adding additional cache often requires a large increase in transistors for a relatively small increase in performance.

But no way is SMT a 50% increase, it's no replacement for additional cores. You can't just create performance out of thin air.

I don't know what Nehalem will bring in terms of performance, but I would expect an IPC gain of maybe 20-30% in applications that utilize the SMT. I'm sure the IMC will increase performance, but Nehalem will also have less cache than Penryn, so that will make for less of an increase.

As for 8 core Nehalem, it'll probably be here on 45nm but only as an extremely high-end solution, and no way will Nehalem be a native octal-core - QC Penryn is already 214mm^2, QC Nehalem would be at least 250mm^2+, no way would native 8-core be feasible. Expect Intel to tape together two quads for a $1,000+ CPU as they did with the QX6700 at launch. You'll need to wait for 32nm for them to become affordable.

 

[Nemesis 1]

BitByBit you got me all excited when you talk about mitosis. I love cells splitting. I have long thought intel was going to introduce mitosis SMT with Net-halem . Its a logical step. It also explains this rush to more cores. Software programmers are struggling with simple single threaded apps. No way is the Desk ready for 16 threads at once. Unless there is a hardware solution to a programming problem . Mitosis could be the ans. IF only it were true. Yes I understand Mitosis is both hardware and software.